Dec 30, 2016 · 28m · 20vc
20VC: How AI Can Enhance Not Replace Humans? What Will Differentiate Between The Winners & The Losers In AI? Why AGI Is Further Away Than We Think with Nitesh Banta, Founder & CEO @ B12
0:00 / 0:00
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of 20VC, host Harry Stebbings interviews Nitesh Banta, Founder and CEO of B12, exploring his transition from venture capital at General Catalyst to tech entrepreneurship. Banta discusses practical strategies for human-assisted artificial intelligence, startup data advantages against monopolies, and realistic timelines for AGI.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Harry holds 24.2% of the talking time here. How this is scored →
Harry as informed peer 2.8 Guest teaching 3.5 Guest disagreement 1.3 Harry pushing back 2.0
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 21:53 Resisting precise AI timeline demands
Nitesh resists Harry's insistence on pinning down specific calendar years for AGI, arguing that narrow AI will dominate for decades and declining to offer simplistic predictions.
Hardest push from Harry ▶ 21:29 Harry presses for hard predictionsHarry refuses to accept generalized statements about AI timelines and directly pushes Nitesh to put concrete numbers and dates on ANI, AGI, and ASI development.
Biggest teaching moment ▶ 14:03 Explaining ML democratization and data edgeNitesh reframes Harry's question about machine learning monopolies by explaining how open-source platforms like TensorFlow democratize models, shifting the real startup competitive edge to proprietary dataset creation.
Harry holds his own ▶ 12:18 Citing x.ai as narrow AI sticky tapeHarry demonstrates technical industry knowledge by citing x.ai's reliance on human checkers to question whether narrow AI applications are largely held together by hidden manual labor.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| Nitesh Banta's Background and Journey to Tech | 1 | 2 | 0 | 0 | Harry opens with friendly introductory questions regarding Nitesh's background. Nitesh walks through his journey from Harvard to Google, GetAround, and General Catalyst without any pushback or tension. | |
| Transitioning from Venture Capital to Founder | 2 | 3 | 2 | 1 | Harry asks if working as a junior VC was a lonely experience. Nitesh gently corrects the premise, noting that General Catalyst had a strong cohort of peers and plenty of human interaction. | |
| Lessons Learned in VC Applied to B12 | 2 | 4 | 0 | 0 | Harry prompts Nitesh on lessons transferred from VC to founding a company. Nitesh delivers an educational summary of the advantages (high deal volume, network) and clear blind spots (lack of operational experience like payroll). | |
| The Reality of AGI vs. the Power of Narrow AI | 3 | 5 | 1 | 1 | Harry brings up Nitesh's belief that AGI is far away. Nitesh educates the audience on narrow AI capabilities versus broad human tasks using a humorous analogy about an AI trying to host Harry's podcast. | |
| Human-in-the-Loop AI and Automation Limits | 5 | 4 | 2 | 5 | Harry offers informed pushback by citing x.ai and asking if narrow AI is mostly held together by 'sticky tape'. He follows up on whether big tech monopolies will dominate machine learning, prompting Nitesh to explain how open-source tools shift startup moats toward unique data sets. | |
| Training Algorithms and Future Work Models | 3 | 4 | 1 | 2 | Harry asks about B12's algorithm training and future work models. Nitesh explains how human-assisted AI eliminates administrative grunt work rather than displacing core roles. | |
| Timeline Predictions for ANI, AGI, and ASI | 4 | 4 | 3 | 6 | Harry forces the issue by demanding hard date predictions for ANI, AGI, and ASI. Nitesh resists committing to precise figures, though Harry presses until Nitesh estimates the 2040s/50s before Harry throws in his own exact prediction. | |
| Quickfire Round: Books, Concerns, and Leadership | 2 | 2 | 1 | 1 | In a standard quickfire round, Nitesh shares book recommendations, his fear that AI overpromises and underdelivers, and leadership hurdles while Harry keeps a quick pace. |
Statements from this episode (16)
Banta: Leaving venture capital becomes economically harder over time
“It gets harder with time. Economically, it gets harder. You get comfortable. You start to feel like you can actually do a good job in the venture industry”
Banta: Venture capital dealmaking is a solo responsibility
“It is a little bit more of a solo sport, because anytime you're doing a deal, you're the person at the firm that's primarily responsible for it.”
Banta Evaluated Over 1,000 Startups Annually as VC Investor
“When I was an investor, I was doing, I was meeting over a thousand companies each year, and I was able to meet with so many other individuals, so that you're just kind of learning from this firehose, and in this unnatural way, starting to learn about new ideas…”
Banta: VC Firms Lack standard Startup OKRs and KPIs
“Venture firms tend to be partnership-oriented. They don't operate around clear OKRs or KPIs. So you're not kind of run in the same efficient manner that traditional startup might be run in.”
Banta: AI startups overpromising human-level AI inevitably disappoint users
“I think a lot of startups often go out, or technology companies go out, and feign this experience that, you know, I'm building this machine or AI that's gonna be able to operate just like a person. And I think one of the challenges there is that you start to c…”
B12 open-sourced Orchestra, a workflow tool integrating human and narrow AI
“One of the things that we're doing at B-Twelve, we open source this tool called Orchestra, which is a human machine based workflow system, and it stitches together narrow pieces of AI with people.”
Banta: Closing the final 10% of AI automation requires most effort
“To get to 80, 90, 95% automation. But the challenge is, like, most people don't care about, you know, if you could get a driverless car that could drive well 85% of the time, that's not particularly useful because that's still too dangerous or inconsistent exp…”
Banta: Human operators behind AI services are not going away soon
“In a lot of cases, there are people behind the machine that are masked or not put at the forefront. And in In certain cases, like, those people aren't going anywhere anytime soon if you want to create the type of experience I think consumers want.”
Banta: Open-source ML tools let junior engineers perform specialist work
“One of the really nice things is that a lot of these companies are open sourcing and democratizing these services, so things like TensorFlow from Google, and the result is, you know, we, when we recruit engineers, I mean, even an out-of-school engineer can do …”
Banta: AI startups build competitive moats through proprietary datasets
“One of the areas where I feel like startups Can have a distinct advantage is creating unique data sets that do interesting things around on top of some of these machine learning algorithms.”
Banta: AI startups gather human data easier than big tech incumbents
“And often as a small company, it's a little bit easier because you don't have The reputation risk and or the challenges that a bigger company might have around organizing a cohort of people that in their world is very small, but in a startup world could be, yo…”
Banta: AI will not replace investors, founders, or marketers near term
“I don't think in the near term you're going to have a machine fully replace the way, you know, venture investors work, fully replace the way startup founders work, podcast producers or marketers.”
Banta: Artificial narrow intelligence phase will continue for decades
“Well, so I think you already see instances where we're firmly in the ANI phase. You see lots of instances where things are self-sufficient. And I think this is a really exciting phase because there are lots of narrow things like driverless cars, as an example,…”
Banta: AGI development is bottlenecked by software complexity, not compute
“I actually don't think the limitation will be compute power, but the limitation will more be around the type of software that you need to build something as complex and dynamic as a person.”
Banta: Artificial super intelligence will follow AGI within weeks to years
“So I feel like once you get to AGI and you're able to have machines that can operate 24 seven, sufficiently work together and do further research in the space, you'll probably get to an ASI in a much shorter period of time. Anything from weeks to months to yea…”